scalable data analysis, modeling, AI, software design at HHMI Janelia
https://mastodon.social/@herrsaalfeld
Stephan Saalfeld
Loading...
New preprint with @lingqiz.bsky.social: Neurodata Without Boredom: Benchmarking Agentic AI for Data Reuse arxiv.org/abs/2605.12808
1/10
1/ 🚨🚨Preprint alert!
"Perfect" isn't always smart 🧠. Anyone can tell apart two stimuli (including our mice). But how do we extract rules that apply to new stimuli? 🧵
biorxiv.org/content/10.6...
@marius10p.bsky.social @computingnature.bsky.social
#neuroAI #neuroscience #AI
Video
Kristin Branson
Nice piece in @theharvardcrimson.bsky.social about our @kempnerinstitute.bsky.social! I say a few things in it.
Miguel Nunez
After 9 years, I am finally finishing the story about normalized cross-correlation (and reviving my blog). The new blog post about deep technical details on the most used image registration technique. I know y'all have been waiting. katpyxa.info/feedbacks/?p...
New paper out! Left-handed DNA cannot bind to natural, right-handed DNA. This leads to lower background if it is used as a probe in DNA-PAINT. Using Fluorigenic DNA-PAINT with left-handed DNA allows for low background, fast 3D DNA-PAINT! Congrats, Bas! skeetorial: 1/x pubs.acs.org/doi/full/10....
Single-molecule localization microscopy (SMLM) techniques offer nanometer-scale resolution by stochastically localizing individual fluorescent molecules. Among these, DNA-mediated point accumulation f...
Hello #world, meet 1,000× Expansion Microscopy.
A small gel would grow to the size of an Olympic swimming pool, while amino-acid-scale distances become visible with ordinary light microscopy.
Led by Helena Hu from @eboyden3.bsky.social's lab, in collab with us.
Story: www.biorxiv.org/content/10.6...
"Warm models showed substantially higher error rates (+10 to +30 percentage points) than their original counterparts, promoting conspiracy theories, providing inaccurate factual information and offering incorrect medical advice. " rdcu.be/fgea6
1/7 New paper accepted as ICML spotlight arxiv.org/abs/2605.03517! We unify self-supervised learning (SSL) algorithms (e.g., contrastive, VICReg, stopgrad) via latent distribution matching (LDM), which matches an induced latent distribution to an explicit latent model.
Really pleased that this work with @dudman.bsky.social is now out in Neuron: A global dopaminergic learning rate enables adaptive foraging across many options
Nature - Experiments on five different language models show that training language models to produce warmer responses can undermine the accuracy of their output, especially when users express...
Harvard’s Kempner Institute made an early bet on owning its own AI computing cluster — a gamble that has quickly made it a rare academic hub for artificial intelligence research, even as private indus...
New release of #BigVolumeBrowser (0.1.1) adds support for (multiscale) OME-Zarr (thanks to libraries from @bogovicj.bsky.social, Stefan Hahmann, Vladimir Ulman).
And midi gizmo navigation (from @herrsaalfeld.bsky.social ), shown below.
Video
Eugene Katrukha 🇺🇦
dlvr.it
Neuron
Grima et al. introduce a novel foraging paradigm to understand how mice can rapidly learn to exploit multi-option environments with varied reward statistics. Decision-making is explained by a model integrating optimal foraging, reinforcement learning, and machine learning. Mesolimbic dopamine best reflects the dynamic global learning rate of the model.
